AI Boom Forces Texas and Beyond to Rethink Energy Supply at Scale

The AI trade has entered that deliciously dangerous phase where everyone keeps whispering about bubbles, but nobody is prepared to lift their foot off the accelerator. And the proof isn’t in NVIDIA’s (NASDAQ:NVDA) price action anymore — it’s in the power grid. If you want to understand where the AI-capex Manhattan Project is heading, stop looking at GPUs and start looking at gigawatts.

As of mid-October, the US data-center pipeline has swollen to an absurd 245 GW, a number so grotesquely oversized it makes 2021-era crypto mining look like a candle flickering in a windstorm. The gravitational pull of that pipeline has shifted decisively toward Texas, where planned capacity has nearly doubled in two quarters as developers stampede into the only geography that can plausibly accommodate this scale of ambition.

The old gospel of “fibre adjacency” has been replaced by a blunt new commandment: access to power is access to survival. And once that truth took hold, the entire map of the industry began to warp. Data Center Alley is giving way to Data Center Prairie — sprawling, isolated, giga-campuses stretching across Pennsylvania steel country, the Wyoming emptiness, and above all, the sun-bleached plains of West and North Texas.

The developers aren’t pretending anymore. They don’t trust the utilities to deliver on time, and they aren’t willing to let grid bottlenecks dictate the pacing of the AI revolution. So they’re doing the unthinkable: building their own power plants, at hyperscaler scale and hyperscaler speed.

Natural gas is the weapon of choice, particularly where the Permian Basin sits practically beneath the server racks. You’re now seeing five-gigawatt campuses in Midland County, two-gigawatt gas-backed parks like Project Horizon, and 5-GW monsters like Pacifico’s Ranch. And yet even the gas network is bursting at capacity, which means if you’re building far from the basin, you’re not just paying for more molecules — you’re paying in time, waiting for new pipelines that may not clear regulatory hurdles before the next economic cycle turns.

Some developers are bending sunlight and wind rather than gas, as in the Nevada and Utah solar-stabilized tract developments or Montana’s Big Sky digital infrastructure campus. But even there, batteries are less about flexibility and more about brute force balancing — smoothing renewables and providing fast-ramping backup for onsite gas turbines. These are no longer clever grid-optimization tricks. They’re survival systems designed for workloads that cannot go dark, ever.

And this all comes with a capital distortion that would make even late-cycle Silicon Valley blush. The 2% of projects with budgets north of USD 17 billion somehow account for 42% of total capital deployment. Meanwhile, the 60% of projects under USD 1 billion barely register — just 8% of total capex. Froth isn’t creeping into the system; it’s flooding the pipes.

The poster children are Project Jupiter in New Mexico at USD 160 billion and Project Kestrel in Missouri at USD 100 billion — numbers so detached from reality they feel like lunar base proposals. What’s more remarkable is the financial engineering underneath them: developers issuing industrial revenue bonds where they are both payer and payee, a circular structure that tees up tax advantages and pushes project valuations into uncharted airspace.

The biggest tell that we’ve entered a new regime is the hyperscalers themselves. For years, they resisted onsite generation, preferring the clean-hands simplicity of grid power. No scope 1 emissions, fewer headaches, shorter contractual lock-ins. But urgency is now trumping orthodoxy. Developers are betting that hyperscalers will overlook the operational risk because speed to compute is paramount, or that a new generation of hyperscalers — less fixated on ESG purity and more obsessed with capacity — will happily sign onto gas-fired campuses if it means teraflops tomorrow instead of teraflops in 2029.

Even though only 10% of the pipeline includes onsite generation, those projects account for a staggering 34% of total capacity. And of course, Texas dominates the map, with gas turbines as the workhorse technology. Every terawatt of this build-out tightens the natural-gas market, competing directly with LNG exports and nudging long-term gas prices higher.

This doesn’t just lift electricity bills for consumers — it crowds utilities out of the turbine market at precisely the moment on-grid demand from EVs, manufacturing reshoring, and electrification is accelerating. If the off-grid AI complexes soak up the turbine supply chain, the regulated utilities will be left trying to keep the lights on with one hand tied behind their back.

And here’s the real kicker: even if the AI bubble burps, pops, or deflates gently, the affordability and reliability pressures unleashed by this shift will not disappear. Once tens of gigawatts of private, gas-fired compute fortresses start pulling fuel, burning molecules, and shaping regional power curves, state regulators are going to step in. First softly, then loudly. And once that happens, all bets are off. Markets can price a bubble; they can’t price a political backlash against runaway energy demand.

This is no longer about whether AI valuations are rich or whether project IRRs make sense on a spreadsheet. The market is staring at an AI-capex universe that is forcibly rewriting America’s energy system in real time. The Manhattan Project metaphor is no longer figurative — it’s literal. The new bottleneck isn’t chips. It’s joules. And every hyperscaler knows it.